Survival prediction models: an introduction to discrete-time modeling

Table 4 Cross-validated time-dependent R-squared measure of Brier score and AUC for the prediction models applied to the pbc test data to predict survival at w=7 years

Method	Type	No. Intervals	R²	AUC
Neural network	Continuous	15	0.409	0.859
RSF	Continuous	-	0.392	0.880
CForest	Discrete	8	0.377	0.869
GBM	Discrete	6	0.363	0.826
Support vector machine	Discrete	5	0.354	0.854
Elastic net	Discrete	25	0.345	0.845
Cox PH	Discrete	-	0.332	0.831
Logistic regression	Discrete	11	0.329	0.831

Higher values of AUC and R² scores indicate better performance. Results are sorted by decreasing R² (best to worst). Type indicates whether the method is applied to continuous- or discrete-time data. AUC: area under the ROC curve; CForest: conditional inference random forest; GBM: gradient boosting machines; PH: proportional hazards; RSF: random survival forest

ISSN: 1471-2288